|
From: Jeremy F. <je...@go...> - 2005-01-10 16:34:19
|
On Mon, 2005-01-10 at 09:45 +0000, Tom Hughes wrote: > Right. I've tried it on a range of systems now. Initial comments > are that it all looks pretty nice but I have a few problems... Download a new copy of the patch; I've fixed some things since the first announce. > - First up, you have used AS_HELP_STRING in configure.in but > none of my older systems know that even though they have > newer autoconfs than they originally shipped with. Only a > problem for people using CVS of course. That seems to be a general autoconf problem. I had to install a new autoconf on my RH7.3 system to get the CVS HEAD to build anyway. > - Second, on RH9 the clone used to start threads is not trapped > because CLONE_SYSVSEM is not set. I just added an extra case > for that to the switch. I changed the switch to only look at flags we really care about. > The big one is that something is causing it to hang. Different systems > hang in different places but all my systems hang at some point during > the regression tests: > > - RH72 and RH73 hang in corecheck/tests/pth_atfork1 > > - RH80 hangs in corecheck/tests/pth_cancel1 > > - RH9 and FC3 hang in none/tests/x86/yield The yield one is a bug in the test. It should be pthread_cond_broadcast, not signal. But the test is bogus now anyway, since it asserts that "rep;nop" has the same scheduling effect as "sched_yield", which definitely isn't true now. J |